postcopy requires to place a whole host page, while migration thread
migrate memory in target page size. This makes postcopy need to collect
all target pages in one host page before placing via userfaultfd.
To enable compress during postcopy, there are two problems to solve:
1. Random order for target page arrival
2. Target pages in one host page arrives without interrupt by target
page from other host page
The first one is handled by previous cleanup patch.
This patch handles the second one by:
1. Flush compress thread for each host page
2. Wait for decompress thread for before placing host page
Signed-off-by: Wei Yang <richardw.yang@linux.intel.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
After using number of target page received to track one host page, we
could have the capability to handle random order target page arrival in
one host page.
This is a preparation for enabling compress during postcopy.
Signed-off-by: Wei Yang <richardw.yang@linux.intel.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
For the first target page, all_zero is set to true for this round check.
After target_pages introduced, we could leverage this variable instead
of checking the address offset.
Signed-off-by: Wei Yang <richardw.yang@linux.intel.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
In postcopy, it requires to place whole host page instead of target
page.
Currently, it relies on the page offset to decide whether this is the
last target page. We also can count the target page number during the
iteration. When the number of target page equals
(host page size / target page size), this means it is the last target
page in the host page.
This is a preparation for non-ordered target page transmission.
Signed-off-by: Wei Yang <richardw.yang@linux.intel.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Compress is not supported with postcopy, it is safe to wait for
decompress thread just in precopy.
This is a preparation for later patch.
Signed-off-by: Wei Yang <richardw.yang@linux.intel.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Usually, incoming migration coroutine yields to the main loop
while its IO-channel is waiting for data to receive. But there is a case
when RAM migration and data receive have the same speed: VM with huge
zeroed RAM. In this case, IO-channel won't read and thus the main loop
is stuck and for instance, it doesn't respond to QMP commands.
For this case, yield periodically, but not too often, so as not to
affect the speed of migration.
Signed-off-by: Yury Kotov <yury-kotov@yandex-team.ru>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
When using hugepages, rate limiting is necessary within each huge
page, since a 1G huge page can take a significant time to send, so
you end up with bursty behaviour.
Fixes: 4c011c37ec ("postcopy: Send whole huge pages")
Reported-by: Lin Ma <LMa@suse.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
ram_save_queue_pages() has an 'err' label that can be replaced by
'return -1' instead.
Same thing with ram_discard_range(), and in this case we can also
get rid of the 'ret' variable and return either '-1' on error
or the result of ram_block_discard_range().
CC: Juan Quintela <quintela@redhat.com>
CC: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
If we are exiting due to an error/finish/.... Just don't try to even
touch the channel with one IO operation.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Fill everything with zero, so the padding fields are also initialized.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Use WITH_RCU_READ_LOCK_GUARD to avoid exiting colo_init_ram_cache
without releasing RCU.
Cc: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
../migration/ram.c: In function ‘multifd_recv_thread’:
/home/elmarco/src/qq/include/qapi/error.h:165:5: error: ‘block’ may be used uninitialized in this function [-Werror=maybe-uninitialized]
165 | error_setg_internal((errp), __FILE__, __LINE__, __func__, \
| ^~~~~~~~~~~~~~~~~~~
../migration/ram.c:818:15: note: ‘block’ was declared here
818 | RAMBlock *block;
| ^~~~~
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
During migration, a tmp page is allocated so that we could place a whole
host page during postcopy.
Currently the page is allocated during load stage, this is a little bit
late. And more important, if we failed to allocate it, the error is not
checked properly. Even it is NULL, we would still use it.
This patch moves the allocation to setup stage and if failed error
message would be printed and caller would notice it.
Signed-off-by: Wei Yang <richardw.yang@linux.intel.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Commit f3f491fcd6 ('Postcopy: Maintain unsentmap') introduced
unsentmap to track not yet sent pages.
This is not necessary since:
* unsentmap is a sub-set of bmap before postcopy start
* unsentmap is the summation of bmap and unsentmap after canonicalizing
This patch just removes it.
Signed-off-by: Wei Yang <richardw.yang@linux.intel.com>
Message-Id: <20190819061843.28642-3-richardw.yang@linux.intel.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
All pages, either partially sent or partially dirty, will be discarded in
postcopy_send_discard_bm_ram(), since we update the unsentmap to be
unsentmap = unsentmap | dirty in ram_postcopy_send_discard_bitmap().
This is not necessary to do discard when canonicalizing bitmap. And by
doing so, we separate the page discard into two individual steps:
* canonicalize bitmap
* discard page
Signed-off-by: Wei Yang <richardw.yang@linux.intel.com>
Message-Id: <20190819061843.28642-2-richardw.yang@linux.intel.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
When encounter error, multifd_send_thread should always notify who pay
attention to it before exit. Otherwise it may block migration_thread
at multifd_send_sync_main forever.
Error as follow:
-------------------------------------------------------------------------------
(gdb) bt
#0 0x00007f4d669dfa0b in do_futex_wait.constprop.1 () from /lib64/libpthread.so.0
#1 0x00007f4d669dfa9f in __new_sem_wait_slow.constprop.0 () from /lib64/libpthread.so.0
#2 0x00007f4d669dfb3b in sem_wait@@GLIBC_2.2.5 () from /lib64/libpthread.so.0
#3 0x0000562ccf0a5614 in qemu_sem_wait (sem=sem@entry=0x562cd1b698e8) at util/qemu-thread-posix.c:319
#4 0x0000562ccecb4752 in multifd_send_sync_main (rs=<optimized out>) at /qemu/migration/ram.c:1099
#5 0x0000562ccecb95f4 in ram_save_iterate (f=0x562cd0ecc000, opaque=<optimized out>) at /qemu/migration/ram.c:3550
#6 0x0000562ccef43c23 in qemu_savevm_state_iterate (f=0x562cd0ecc000, postcopy=false) at migration/savevm.c:1189
#7 0x0000562ccef3dcf3 in migration_iteration_run (s=0x562cd09fabf0) at migration/migration.c:3131
#8 migration_thread (opaque=opaque@entry=0x562cd09fabf0) at migration/migration.c:3258
#9 0x0000562ccf0a4c26 in qemu_thread_start (args=<optimized out>) at util/qemu-thread-posix.c:502
#10 0x00007f4d669d9e25 in start_thread () from /lib64/libpthread.so.0
#11 0x00007f4d6670635d in clone () from /lib64/libc.so.6
(gdb) f 4
#4 0x0000562ccecb4752 in multifd_send_sync_main (rs=<optimized out>) at /qemu/migration/ram.c:1099
1099 qemu_sem_wait(&p->sem_sync);
(gdb) list
1094 }
1095 for (i = 0; i < migrate_multifd_channels(); i++) {
1096 MultiFDSendParams *p = &multifd_send_state->params[i];
1097
1098 trace_multifd_send_sync_main_wait(p->id);
1099 qemu_sem_wait(&p->sem_sync);
1100 }
1101 trace_multifd_send_sync_main(multifd_send_state->packet_num);
1102 }
1103
(gdb) p i
$1 = 0
(gdb) p multifd_send_state->params[0].pending_job
$2 = 2 //It means the job before MULTIFD_FLAG_SYNC has already fail
(gdb) p multifd_send_state->params[0].quit
$3 = true
Signed-off-by: Ivan Ren <ivanren@tencent.com>
Message-Id: <1567044996-2362-1-git-send-email-ivanren@tencent.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
There is a race between TCG and accesses to the dirty log:
vCPU thread reader thread
----------------------- -----------------------
TLB check -> slow path
notdirty_mem_write
write to RAM
set dirty flag
clear dirty flag
TLB check -> fast path
read memory
write to RAM
Fortunately, in order to fix it, no change is required to the
vCPU thread. However, the reader thread must delay the read after
the vCPU thread has finished the write. This can be approximated
conservatively by run_on_cpu, which waits for the end of the current
translation block.
A similar technique is used by KVM, which has to do a synchronous TLB
flush after doing a test-and-clear of the dirty-page flags.
Reported-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The purpose of the calculation is to find a HostPage which is partially
dirty.
* fixup_start_addr points to the start of the HostPage to discard
* run_start points to the next HostPage to check
While in the middle stage, there would two cases for run_start:
* aligned with HostPage means this is not partially dirty
* not aligned means this is partially dirty
When it is aligned, no work and calculation is necessary. run_start
already points to the start of next HostPage and is ready to continue.
When it is not aligned, the calculation could be simplified with:
* fixup_start_addr = QEMU_ALIGN_DOWN(run_start, host_ratio)
* run_start = QEMU_ALIGN_UP(run_start, host_ratio)
By doing so, run_start always points to the next HostPage to check.
fixup_start_addr always points to the HostPage to discard.
Signed-off-by: Wei Yang <richardw.yang@linux.intel.com>
Message-Id: <20190806004648.8659-2-richardw.yang@linux.intel.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
In postcopy-ram.c, we provide three functions to discard certain
RAMBlock range:
* postcopy_discard_send_init()
* postcopy_discard_send_range()
* postcopy_discard_send_finish()
Currently, we allocate/deallocate PostcopyDiscardState for each RAMBlock
on sending discard information to destination. This is not necessary and
the same data area could be reused for each RAMBlock.
This patch defines PostcopyDiscardState a static variable. By doing so:
1) avoid memory allocation and deallocation to the system
2) avoid potential failure of memory allocation
3) hide some details for their users
Signed-off-by: Wei Yang <richardw.yang@linux.intel.com>
Message-Id: <20190724010721.2146-1-richardw.yang@linux.intel.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
When migrate_cancel a multifd migration, if run sequence like this:
[source] [destination]
multifd_send_sync_main[finish]
multifd_recv_thread wait &p->sem_sync
shutdown to_dst_file
detect error from_src_file
send RAM_SAVE_FLAG_EOS[fail] [no chance to run multifd_recv_sync_main]
multifd_load_cleanup
join multifd receive thread forever
will lead destination qemu hung at following stack:
pthread_join
qemu_thread_join
multifd_load_cleanup
process_incoming_migration_co
coroutine_trampoline
Signed-off-by: Ivan Ren <ivanren@tencent.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Message-Id: <1561468699-9819-4-git-send-email-ivanren@tencent.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
When we 'migrate_cancel' a multifd migration, live_migration thread may
hung forever at some points, because of multifd_send_thread has already
exit for socket error:
1. multifd_send_pages may hung at qemu_sem_wait(&multifd_send_state->
channels_ready)
2. multifd_send_sync_main my hung at qemu_sem_wait(&multifd_send_state->
sem_sync)
Signed-off-by: Ivan Ren <ivanren@tencent.com>
Message-Id: <1561468699-9819-3-git-send-email-ivanren@tencent.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
---
Remove spurious not needed bits
When we 'migrate_cancel' a multifd migration, live_migration thread may
go into endless loop in multifd_send_pages functions.
Reproduce steps:
(qemu) migrate_set_capability multifd on
(qemu) migrate -d url
(qemu) [wait a while]
(qemu) migrate_cancel
Then may get live_migration 100% cpu usage in following stack:
pthread_mutex_lock
qemu_mutex_lock_impl
multifd_send_pages
multifd_queue_page
ram_save_multifd_page
ram_save_target_page
ram_save_host_page
ram_find_and_save_block
ram_find_and_save_block
ram_save_iterate
qemu_savevm_state_iterate
migration_iteration_run
migration_thread
qemu_thread_start
start_thread
clone
Signed-off-by: Ivan Ren <ivanren@tencent.com>
Message-Id: <1561468699-9819-2-git-send-email-ivanren@tencent.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reproduce the problem:
migrate
migrate_cancel
migrate
Error happen for memory migration
The reason as follows:
1. qemu start, ram_list.dirty_memory[DIRTY_MEMORY_MIGRATION] all set to
1 by a series of cpu_physical_memory_set_dirty_range
2. migration start:ram_init_bitmaps
- memory_global_dirty_log_start: begin log diry
- memory_global_dirty_log_sync: sync dirty bitmap to
ram_list.dirty_memory[DIRTY_MEMORY_MIGRATION]
- migration_bitmap_sync_range: sync ram_list.
dirty_memory[DIRTY_MEMORY_MIGRATION] to RAMBlock.bmap
and ram_list.dirty_memory[DIRTY_MEMORY_MIGRATION] is set to zero
3. migration data...
4. migrate_cancel, will stop log dirty
5. migration start:ram_init_bitmaps
- memory_global_dirty_log_start: begin log diry
- memory_global_dirty_log_sync: sync dirty bitmap to
ram_list.dirty_memory[DIRTY_MEMORY_MIGRATION]
- migration_bitmap_sync_range: sync ram_list.
dirty_memory[DIRTY_MEMORY_MIGRATION] to RAMBlock.bmap
and ram_list.dirty_memory[DIRTY_MEMORY_MIGRATION] is set to zero
Here RAMBlock.bmap only have new logged dirty pages, don't contain
the whole guest pages.
Signed-off-by: Ivan Ren <ivanren@tencent.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Message-Id: <1563115879-2715-1-git-send-email-ivanren@tencent.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
